Exploratory Analysis of Suburban Land Cover
and Population Density in the U.S.A.
Francesca Pozzi 1 and Christopher Small 2
1
CIESIN
Columbia University
Palisades, NY 10964 USA
[email protected]
2
Lamont Doherty Earth Observatory
Columbia University
Palisades, NY 10964 USA
[email protected]
Reprinted from:
IEEE/ISPRS joint Workshop on Remote Sensing and Data Fusion over Urban Areas,
Paper 35
Rome, Italy
8-9 November, 2001
Copyright © 2001 IEEE. Reprinted from IEEE/ISPRS joint Workshop on Remote
Sensing and Data Fusion over Urban Areas. This material is posted here with permission of
the IEEE. Internal or personal use of this material is permitted. However, permission to
reprint/republish this material for advertising or promotional purposes or for creating new
collective works for resale or redistribution must be obtained from the IEEE by sending a
blank email message to [email protected]. By choosing to view this document,
you agree to all provisions of the copyright laws protecting it.
1
Exploratory Analysis of Suburban Land Cover and
Population Density in the U.S.A.
Francesca Pozzi, Christopher Small
Abstract— The objective of this study is to investigate the
consistency of “suburban” population densities and land
covers. We analyzed population density, extracted from
the census, and vegetation abundance, derived from Landsat imagery, taking six cities in the U.S.A. as contrasting
examples. Combining population density and areal vegetation abundance estimates yields univariate and bivariate
distributions for the two variables. We quantify the relationship between population density and vegetation fraction
in Atlanta, Chicago, Los Angeles, New York, Phoenix and
Seattle. A bimodal distribution of population density in
the U.S.A. suggests that it may be possible to characterize
“suburban” areas on the basis of population density between
100 and 10,000 people/km2 . The maximum areal vegetation
cover diminishes linearly with the Log10 of population density in cities with large density ranges.
Keywords— Suburban, Population Density, Vegetation
Fraction, Land Cover
I. Introduction
Suburban areas in the U.S.A.are often perceived as the
greener “residential areas on the outskirts of a city or a
large town” [1], socially and economically dependent on
large cities. Suburbs have been given consideration in the
past decade, given the close connections with cities and
the negative consequences associated with urban sprawl.
These consequences include loss of agricultural land and
natural vegetation, increased traffic congestion and associated degradation of air quality.
According to the U.S.Census Bureau [2], between 1995
and 1996 more than 2 million people moved from U.S.A.
cities and from non-metropolitan areas into the “suburbs”.
In the same report suburbs are defined as “all territory
within an Metropolitan Statistical Areas (MSA) but outside of a central city”. Although this definition is intuitive
and easily understandable, there appears to be no consistent or formal characterization of suburban areas in terms
of physical or socioeconomic characteristics.
In recent years, many authors have considered the relationship between population characteristics and environmental variables, but their interests and goals are different
from the ones presented in this paper. Examples are [3],
[4], [5], where the authors presented studies of integration
of population and land cover for quality of life assessment
or improvement of land cover classification. Studies [6]
and [7] considered population and vegetation to better understand urban dynamics. Other authors analyzed land
cover change related to population change and its impacts
F. Pozzi is a research associate at the Center for International Earth
Science Information Network (CIESIN) of Columbia University. Email: [email protected]
C. Small is a researcher at the Lamont Doherty Earth Observatory
of Columbia University. E-mail: [email protected]
on the surrounding natural areas [8]. Approaches more focused on the correlation between socioeconomic variables,
such as population count and housing density, and land
cover are presented in [9] and [10].
The above mentioned studies focused mainly on the integration of population and land cover data and on single
case studies. To our knowledge, no study has yet been performed to examine the demographic and land cover characteristics of suburban areas and their consistency across
different physiographic environments.
The purpose of this paper is to investigate the question
of whether suburban areas can be defined based on demographic and physical characteristics, specifically population
density and vegetation cover. Based on the expectation
that suburban areas are greener than urban centers, and
that the predominant suburban land cover is vegetation,
we attempt to quantify the extent to which suburban areas are vegetated in different U.S.A. cities. We consider
suburban areas based on population density and on apparent spectral reflectance using Landsat data to quantify the
relationship between the two variables looking at the cities
of Atlanta, Chicago, Los Angeles, New York, Phoenix and
Seattle.
II. Data
Given the purpose of the study, the six cities we considered present very different characteristics, in terms of geographic location and spatial structure and dynamics. We
chose cities located in a temperate climate, both in a deciduous forest biome (New York, Chicago, Atlanta) and in
an evergreen forest biome (Seattle) and cities located in an
arid or semi-arid climate (Los Angeles and Phoenix). We
also included cities that have been among the most fastgrowing of the past decades in the U.S.A. (Phoenix and
Seattle), and cities that have experienced rapid growth in
the past and now are characterized by large population
(New York, Chicago, Los Angeles).
A. Population Density
We calculated population density from the 1990 U.S.
Census Bureau population counts at the block level, the
lowest in the U.S. census structural hierarchy. These data
are available separately as spatial data (Topologically Integrated Geographic Encoding and Referencing system TIGERr) and tabular data (Summary Tape Files-STFs)
for each county in the U.S.A.. In this study we considered
population density, expressed in persons/km2 . For each
city we selected one or more counties containing the Centered Business District (CBD) and the surrounding sub-
2
urbs. This resulted in the selection of the following counties, for which we then created a smaller subset concordant
with Landsat coverage (area reported in parenthesis):
2
• Atlanta: DeKalb and Fulton (900 km );
2
• Chicago: Cook (950 km );
2
• Los Angeles: Los Angeles (3100 km );
• New York Metropolitan Area: Bronx, Kings, New
York, Queens, Richmond, Bergen, Essex, Hudson, Passaic,
Nassau, Rockland, Westchester (2000 km2 );
2
• Phoenix: Maricopa (4700 km );
2
• Seattle: King (3200 km ).
B. Vegetation Fraction
The characteristic spatial scale and the spectral variability of urban and suburban land cover poses serious problems for traditional image classification algorithms. In areas where the reflectance spectra of the land cover vary
appreciably at scales comparable to, or smaller than, the
Ground Instantaneous Field Of View (GIFOV) of most
satellite sensors, the spectral reflectance of a individual
pixel will generally not resemble the reflectance of a single
land cover class but rather a mixture of the reflectances of
two or more classes present within the GIFOV. Because
they are combinations of spectrally distinct land cover
types, mixed pixels in urban areas are frequently misclassified as other land cover classes. Similarly, the definition
of an “urban” spectral class will usually incorporate pixels
of other non-urban classes.
If an urban area contains significant amounts of vegetation then the reflectance spectra measured by the sensor will be influenced by the reflectance characteristics of
the vegetation. Macroscopic combinations of homogeneous
“endmember” materials within the GIFOV produce a composite reflectance spectrum that can often be described as
a linear combination of the spectra of the endmembers [11].
If mixing between the endmember spectra is predominantly
linear and the endmembers are known a priori, it may be
possible to “unmix” individual pixels by estimating the
fraction of each endmember in the composite reflectance
of a mixed pixel [12], [13].
Analysis of Landsat TM imagery suggests that the spectral reflectance of many urban areas can be described as
linear mixing of three distinct spectral endmembers [14],
[15]. Principal component analysis of urban reflectance
consistently yield eigenvalue distributions suggesting that
the majority of scene variance is contained within a two dimensional mixing plane. The triangular distribution in the
mixing space defined by the principal components bears a
similarity to the well known Tasseled Cap distribution discovered by [16]. The feature space distributions are similar
in the sense that both contain a vegetation endmember
that is distinct from a mixing continuum between high and
low albedo endmembers.
The spectral endmembers determined for the areas investigated here correspond to low albedo (e.g. water, shadow,
roofing), high albedo (e.g. cloud, sand, roofing) and vegetation. The strong visible absorption and infrared reflectance that is characteristic of vegetation is sufficiently
distinct from the spectrally flat reflectance of the low and
high albedo endmembers to allow the three components to
be “unmixed” by inverting a simple three component linear mixing model [14]. The result of the unmixing is a
set of fraction images showing the areal percentages, given
as fractions between 0 and 1, of each endmember present
within each pixel. Analysis of Landsat, Ikonos and AVIRIS
imagery of several urban/suburban areas shows that a
three component linear mixing model provides stable, consistent estimates of vegetation fraction for both constrained
and unconstrained inversions using three different endmember selection methods [15]. Vegetation fraction estimates
derived from Landsat TM data were validated with aeral
vegetation fractions calculated from 2 m aerial photography and generally showed agreement to within 10% [14]
The vegetation fraction estimates given here were derived
from Landsat TM and validated with Ikonos MSI imagery.
III. Analysis and Results
The first step of the analysis was to quantify the distribution of population density across the entire United States
to estimate whether rural, urban and suburban areas are
clearly discernible based on population density. The distribution of people as a function of population density for
the U.S.A. in 1990 is bimodal (Figure 1) [17]. The modes
represent the spatially concentrated settlements near cities
and the spatially dispersed settlements farther from cities.
The larger mode has a distinct break in slope near 10,000
people/km2 , and a short, high density tail. This tail corresponds to the high density cores of large cities. For the purposes of this study we consider suburban areas to be characterized by population density between 100 and 10,000
people/km2 .
To perform the study on the six cities, spatial and tabular data from the Census were initially aggregated based
on the block numeric codes for each county. The resulting vector layers were then projected to UTM coordinates,
Fig. 1. Histogram of the Population Density for the U.S.A., showing
also the distribution for Eastern U.S.A. (East of the 90 ◦ W, black
line) and for Western U.S.A.(grey line).
3
Fig. 2. Population Density and Vegetation Fraction at full resolution for part of New York Metropolitan Area. Black areas represent no
data for population density and no vegetation (water) for the vegetation fraction.
rasterized to a 30 m grid and coregistered to the Landsat
data. To quantify the relationship between population density and vegetation fraction, we produced bivariate distributions of people and land area as functions of population
density and vegetation fraction. The bivariate population
distributions are shown with population density and vegetation fraction for each city in Figure 3. We then summed
the bivariate distributions to produce marginal distributions of people as functions of population density and vegetation fraction for each city (Figure 4).
IV. Discussion
In the six cities we studied there appears to be a pattern
for suburban areas, both in terms of population density and
vegetation fractions. The population density histograms,
calculated for the counties listed in II-A, show the suburban peak characteristic of the entire U.S.A. with Atlanta
and New York at the extremes (Figure 4. The vegetation
fraction histograms also show a consistent pattern, with
peaks varying between about 0.1 and 0.55. The cities with
large urban core also present a smaller peak for vegetation
fractions less than 0.01. Prominent exceptions are Atlanta,
which has a symmetric distribution centered on higher values (0.5 to 0.6) and New York, which has a long tailed
monotonic distribution, with a peak at less than 0.1.
The differences between the physiographic environments
and the urban structures for the six cities are such that
the peaks of the bivariate histograms are spread across
a range of population densities and vegetation fractions.
Nonetheless a consistent sub-linear relationship is seen for
the largest cities (New York, Chicago and Los Angeles).
These three cities have similar density distributions, with
comparable peak values and with vegetation fractions linearly decreasing with Log10 (Population Density). Phoenix
and Seattle have the most similar population density distributions, but their vegetation fraction distributions are
different, due to their arid and humid climates. The bivariate distributions for these two cities are more isotropic
than the larger cities, with Phoenix containing large inhabited unvegetated areas and Seattle characterized by large
uninhabited and densely vegetated areas. Atlanta, on the
other hand, presents a more uniform distribution, with little variations in either vegetation fraction and population
density.
V. Conclusions
The objective of this study was to investigate the consistency of “suburban” settlement patterns, based on the relationship between population density and land cover among
different cities in the U.S.A.. The principal conclusion of
this study is that the population density distribution in the
U.S.A. may provide a demographic basis for distinguishing
urban, suburban and rural areas. The block level population density distribution for the entire U.S.A. shows two
distinct modes corresponding to moderate density (100 to
10,000 people/km2 ) settlements surrounding higher density urban cores and to low density settlements (less than
100 people/km2 ) dispersed throughout the country. The
high density (more than 10,000 people/km2 ) cores correspond to a distinct tail delineated by a break in slope on
the moderate density ”suburban” mode. This demographic
classification would place 71% of Americans in suburban
areas, 25% in rural areas and 3% in urban areas in 1990.
The wide variety of land use classes that character-
4
Fig. 3. Spatial Distributions of Population and Vegetation. Combining population density with vegetation fraction yields a demographic
classification, where rural population densities are shown in blue, urban in red and suburban in green. Different shades of green correspond
to different amounts of vegetation. Note the similarity of the peaks in the bivariate distributions for Chicago, New York and Los Angeles.
Full resolution images are available at www.LDEO.columbia.edu/˜ small/Urban.html
5
and distribution also control evapotranspiration and albedo
thereby influencing climate dynamics at regional and local
scales. Coanalysis of settlement patterns and land cover
characteristics may eventually facilitate quantitative analysis of urban sprawl, natural resource management and
land use policy implications, when the relationship between
these factors is understood.
Aknowledgements
The authors would like to acknowledge the support and
resources of CIESIN’s SocioEconomic Data and Application Center (SEDAC), the Columbia Earth Institute and
Lamont-Doherty Earth Observatory of Columbia University.
References
[1]
[2]
[3]
[4]
[5]
Fig. 4. Univariate Distributions of number of people as a function of
Population Density (top) and Vegetation Fraction for the six cities.
[6]
[7]
ize urban and suburban areas poses serious problems for
thematic classifications that rely on moderate resolution
imagery. Unlike most other landcover classes, the urban/suburban mosaic is consistent only in its spectral heterogeneity at the scale of most operational satellite sensor GIFOVs [15]. The resolution difference between census
tracts and the Landsat sensor does not allow for a simple
spectral characterization of suburban landcover.
In the U.S.A. cities we investigated, the most consistent
spectral characteristic of “demographically suburban” areas was related to the amount of vegetation cover. Maximum vegetation fraction generally diminishes with increasing population density but spectral heterogeneity still results in a wide range of vegetation fractions in demographically suburban areas. Large cities with high density urban cores do, however, show a distinct linear decrease in the modal vegetation fraction with increasing
Log10 (population density). The different physical environments of the cities considered here result in different vegetation distributions at different settlement densities. We
find no evidence for a single consistent relationship between
suburban population density and vegetation abundance in
the U.S.A..
Quantitative characterization of vegetation abundance in
suburban areas does however provide a basis for comparison of the physical environments in which most Americans
reside. Vegetation has a direct impact on solar energy flux
through the environment and therefore influences the microclimate of the human habitat. Vegetation abundance
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Merriam-Webster Dictionary, Merriam-Webster, 2001.
Hansen K.A., “Geographic Mobility: March 1995 to March
1996,” Tech. Rep., U.S.Census Bureau, November 1997.
Lo P. and Faber B.J., “Integration of Landsat Thematic Mapper and Census Data for Quality of Life Assessment,” Remote
Sensing of Environment, vol. 2, pp. 143–157, 1997.
Harris P. M. and Ventura S. J., “The Integration of Geographic
Data with Remotely Sensed Imagery to Improve Classification
in an Urban Area,” Photogrammetric Engineering and Remote
Sensing, vol. 61, no. 8, pp. 993–998, 1995.
Vogelmann J. E. et al, “Regional Characterization of Land Cover
Using Multiple Sources of Data,” Photogrammetric Engineering
and Remote Sensing, vol. 64, no. 1, pp. 45–57, 1998.
Emmanuel R., “Urban Vegetational Change as Indicator of Demographic Trends in Cities: the Case of Detroit,” Environment
and Planning, vol. B 24, pp. 415–426, 1997.
Wagner T.W.and Ryznar R.M., “Toward Understanding Urban
Processes with Remotely Sensed Data: Examples from Detroit,”
in Proceeding of GeoInformatics’99 Conference, University of
Michigan, Ann Arbor, MI, 1999.
Wang Y. and Zhang X., “Land Cover Change of Metropolitan
Chicago Area from 1972 to 1997 and the Impact to Natural Communities in the Region,” in Proceedings of GeoInformatics’99,
University of Michigan, Ann Arbor, MI, 1999.
Smith R.M. Yuan Y. and Limp W.F., “Remodeling Census Population with Spatial Information form Landsat TM Imagery,”
Computers, Environment and Urban Systems, vol. 21, no. 3/4,
pp. 245–258, 1997.
Radeloff V. C. et al, “Exploring the Spatial Relationship Between Census and Land-Cover Data,” Society and Natural Resources, vol. 13, no. 6, pp. 599–609, 2000.
Singer R. B. and McCord T.B., “Large Scale Mixing of Bright
and Dark Surface Materials and Implications for Analysis of
Spectral Reflectance,” Journal of Geophisical Research, pp.
1835–1848, 1979.
Adams J. B. et al, “Spectral Mixture Modeling: a New Analysis
of Rock and Soil Types at the Viking Lander 1 Site,” Journal
of Geophysical Research, vol. 91, pp. 8098–8122, 1986.
Boardman J. W., “Inversion of Imaging Spectrometry Data Using Singular Value Decomposition,” in Proceeding of the 12th
Canadian Symposium on Remote Sensing, 1989, pp. 2069–2072.
Small C., “Estimation of Urban Vegetation Abundance by Spectral Mixture Analysis,” International Journal of Remote Sensing, vol. 22, no. 7, pp. 1305–1334, 2001.
Small C., “Multiresolution Analysis of Urban Reflectance,” in
Proceedings of IEEE joint Workshop on Remote Sensing and
Data Fusion over Urban Areas, Rome, Italy, 8-9 November 2001.
Kauth R. J. and G. S. Thomas, “The Tasseled Cap - a graphic
description of the spectral-temporal development of agricultural
crops as seen by Landsat.,” in Proceedings of the Symposium on
Machine Processing of Remotely Sensed Data, 1976, p. 4041.
Small C., “A Global Analysis of Urban Population Distribution and the Physical Environment,” in Proceedings of the
Open Meeting of the Human Dimension of Global Environmental Change Research Community, Brazil Academy of Science,
Rio de Janeiro, Brazil, 6-8 October 2001.